What is a context window, and why does it matter?
A context window is the amount of text an AI can keep in view at one time. Everything the AI should take into account for an answer has to fit inside it: your instructions, the conversation so far and any documents. What does not fit is simply not seen. This is why long chats lose the thread and why large files are only partly read.
Think of a desk, not a memory
Picture a desk of a fixed size. The AI can read and use whatever lies on the desk. When the desk is full and a new paper arrives, an old one has to come off.
This is different from what the AI learned when it was built. That gave it general knowledge of language and the world. It did not include your price list, your customers or last week's emails. Anything about your company has to be put on the desk each time.
An AI also keeps nothing between two questions. A chat feels like one continuous conversation only because the app puts the whole conversation back on the desk with every new message.
How large is a context window?
It depends on the AI model, and sizes have grown quickly. The size is measured in tokens, the small pieces of text an AI counts in. Today's models can hold anything from a long report to several books.
Four signs that the desk is full, and what helps
| What you notice | Why it happens | What helps |
|---|---|---|
| After a long chat, the AI forgets a rule you gave at the start | Early messages were dropped to make room | Start a new chat and repeat the rule |
| Only part of a large file is used | The file did not fit | Send only the section that matters |
| Answers get slower and cost more | Everything on the desk is read again for every answer | Keep each chat to one subject |
| A detail in the middle of a long text is missed | A fact gets lost among a lot of other text | Give less text |
Is a bigger window the answer?
Partly. A large window lets you hand over a full contract in one go, which is useful. You still have to choose what goes on the desk, for three reasons.
- Everything on the desk is read each time. Cost and waiting time grow with it.
- The more unrelated text surrounds a fact, the easier it is for the AI to overlook it or mix it up with something similar.
- No window holds a whole company. Years of emails, files and records will always be larger.
The aim is to put the right few pages on the desk, not the most pages.
What good AI software does for you
You should not have to manage the desk by hand. Good AI software does three things.
- It searches first. A search finds the passages that fit your question, and only those go to the AI.
- It sums up. When a conversation gets long, the older parts are shortened to a brief summary.
- It starts with what is in front of you. The AI begins with the customer or order you have open, not with everything it could reach.
When you compare AI products, ask how each one handles these three points.
How Worknano handles this
Worknano does all three. The AI Assistant already knows the screen you are on, so on the CRM page it starts with the deal in front of you. To find anything else it uses semantic search, a search that finds text by its meaning. It reads only the passages that matter instead of whole documents, including text inside PDF, Word, Excel and PowerPoint files. Long conversations are shortened by the Assistant Memory Summarizer, so the assistant keeps the thread. And the + button starts a fresh chat when the subject changes.
- Semantic search: find your work by meaning
- AI assistants: the chat that knows the page you are on
- Assistant Memory Summarizer
Create a free workspace and open the assistant on a record to see what it already knows.